Carrier-Grade MPLS: Some Big Missing Pieces
Do carriers really trust IP/MPLS networks to carry their premium services? That's one of the fundamental questions we address in Heavy Reading's latest report, "Resilience, Reliability, and OAM in Converged Networks: A Heavy Reading Competitive Analysis."
The answer is not nearly as clear as the question. The carrier technologists I interviewed for this report essentially want to believe that IP/MPLS technology is resilient and reliable enough to meet their rigorous standards. But they also pointed to the areas where they still have concerns about IP/MPLS. And those concerns are not trivial.
First let's look at MPLS hardware reliability. Carriers feel that there is absolutely no reason, other than bad design, that a piece of IP/MPLS gear shouldn't be "carrier-grade" today. Most vendors are in fact very close to achieving this goal. Telecom networking, however, is neither horseshoes nor hand grenades, so "very close" isn't quite close enough. As vendors add carrier-grade hardware features to their boxes, they also are increasing prices dramatically compared with "enterprise-grade" boxes. But if the resulting platform is still only "almost carrier-grade," carriers are still obliged to install boxes in pairs (in the worst cases) to ensure "five-nines" reliability – only now they're paying a carrier-grade premium to do so.
That's a gloomy way to look at things, but there is a positive gloss: Our product analysis shows that most vendors have made tremendous progress toward true five-nines (or better) hardware reliability.
But what about software? In legacy networks, most of the service creation and maintenance intelligence sits in the network operations center (NOC). The folks in the NOC use their network management terminals to create services across the network and to diagnose and hopefully repair problems as they occur. But converged networks are far more dependent on distributed intelligence, using protocols such as OSPF, BGP, LDP, and RSVP-TE. These protocols will allow carriers to create far more flexible and powerful services on a larger scale than if they were obliged to run everything from the NOC.
The tradeoff for this flexibility is a greater vulnerability to problems caused by protocol software and the embedded operating systems that run it. Let's face it, no software can ever be bug-free, and unfortunately there is no NEBS certification for network operating systems and software. (In fact, there isn't a good NEBS regimen for MPLS at all, as I explain in the report.)
Many of the carriers I interviewed this report were cautiously optimistic that most software problems were at least solvable, although they all felt that the Holy Grail for carrier-grade MPLS software is the ability to perform software upgrades without interrupting existing services that are running on the box. Others were slightly more concerned about today's software. They were less confident that router operating systems were as stable as they could be. They expressed concerns about the ease in which a carrier network can be attacked, either by accident or design. And some were downright worried that the IETF approach to protocol development is too vendor-centric and doesn't have enough representation by carriers.
Perhaps the biggest concern shared by carriers is the lack of effective operations, administration, and maintenance (OAM) protocols for MPLS. IP has never had particularly good OAM capabilities (many IP fans, for example, consider Ping to be an OAM protocol). But in the past, nobody needed IP to have good OAM. After all, the Internet is just a best-effort service, and it's even designed to heal itself if things go wrong. But carriers today are expecting to make money from IP-based services, and a good OAM capability is essential part of controlling operating expenses. Without OAM for MPLS, it's almost certain that carriers will continue to experience poor margins on IP-based services, because their opex will be higher.
Carriers tried to take these requirements to the vendor-dominated IETF two years ago but were turned away. Today, some of those vendors have finally remembered that the customer is always right, and the scramble to create a workable MPLS OAM is on.
Overall, the outlook for IP/MPLS infrastructure is very rosy right now. MPLS-based Layer 3 virtual private networks are this year's hot service. All the building blocks are in place for carriers to deploy Layer 3 VPNs, and at the prices being offered today, there's no shortage of customers. However, it remains to be seen if carriers can really make a sustainable margin from these new services. The reliability, resilience, and OAM capabilities of the MPLS products they choose to install will have a huge effect on service profitability. Right now, carrier success with Layer 3 VPNs is far from a sure thing.
— Geoff Bennett, Chief Technologist, Heavy Reading